SETF clarification

[JONL at MIT-MC (Jon L White)]

I believe the first LISPM implementations of SETF were explicitly
described to have "undefined" value -- mostly because of the random
nature of the code produced by the SETF expanders.  While it might
be desirable to standardize someday on a useful value, I'd be curious
to hear of anyone who has much code written depending on the return
value from SETF.  The BMT incident shows that likely he would get
varying results when running such code.